Odd TPF 2p on regular smp

[Patonb]

So thanks to the 8101 my 2p l5639 machine is less useful than my 950 machine.

Has anyone got any ideas why I'm getting load imbalances of 17-48% with a3, while a5 were approx 1%?

I'm using a Vbox install of Ubuntu.
Yes to kraken, bfs.
Top shows thekraken running

I was pulling 23 min tpf on 6900/1, 29min tpf on 6903, 69min tpf on 6904s, I was pulling 49min tpf on the 8101.

My 950 use to do 28min tpf on 6900/1, which is whats peeving me.. The 2p can do the BA faster, yet regular smp its behind.

Not a happy camper right now

 

[extide]

What tpfs are you getting in SMP?

 

[Patonb]

Forgot that, sorry...

Currently on 6947

3:30ish, its down to 3:03 right now.

Hmm, it seems to be dropping, as doing the math, its 2:59

My 950 @4.0Ghz is getting 2:54

 

[Nathan_P]

Update to the latest version of the kraken, v0.6 will restart the core after 15mins to try and get DLB to engage. I've had this problem on my rig and its always caused by DLB not engaging.

 

[firedfly]

Update to the latest version of the kraken, v0.6 will restart the core after 15mins to try and get DLB to engage. I've had this problem on my rig and its always caused by DLB not engaging.

AFAIK, only the larger bigadv units (p6903, p6904, p8101) have a good success rate of engaging DLB after resuming from a checkpoint. It is pointless to restart when running -smp only.

 

[Patonb]

I am running thekraken v6

I noticed that the unit didnt restart after my 1st check point was done.

Grrrrrrrrr, not a happy camper, 6k ppd and hope for a 690x BA, or 20k ppd on a3.

 

[Untitledone]

AFAIK, only the larger bigadv units (p6903, p6904, p8101) have a good success rate of engaging DLB after resuming from a checkpoint. It is pointless to restart when running -smp only.

I think it depends on the system. The 8P almost always engages DLB, from the start, and after kraken auto-restart. So DLB engages 2x per unit, so 30-40 times per day when running SMP balls out lol.

 

[firedfly]

I think it depends on the system. The 8P almost always engages DLB, from the start, and after kraken auto-restart. So DLB engages 2x per unit, so 30-40 times per day when running SMP balls out lol.

In that case, you shouldn't have the kraken set up to auto-restart.

 

[Grandpa_01]

I noticed last night that DLB starts almost immediately most of the time on my 4P's when running smp

 

[Untitledone]

In that case, you shouldn't have the kraken set up to auto-restart.

Haha, maybe, but it does work. Most units need it though. I would rather have it than not, and most of the time in it still engages DLB so it is not really hurting me .

 

[Patonb]

DBL Is kicking in now, with 2-8% imbalance usually, which is back normal, but still crap points for what i was expecting off the BA times of the 2 systems.

 

[Nathan_P]

DBL Is kicking in now, with 2-8% imbalance usually, which is back normal, but still crap points for what i was expecting off the BA times of the 2 systems.

My 5640's:-

6901, 70k PPD
6903/6904 110k PPD

Stock SMP, Max 53k PPD. Thats roughly half depending on project. It's usualyy runing mid 40's

 

[Patonb]

I'm getting 20K ppd max! Yes you are 500mhz faster, but I was expecting about 30-35k.

 

[Nathan_P]

I'm getting 20K ppd max! Yes you are 500mhz faster, but I was expecting about 30-35k.

Something weird is going on with your set up causing it to slow down. What do you get if you run it in native windows with the v7 client??

I know it shouldn't make any difference but i get A4 wu exclusively on my win 7 box and A3 only on my v6 linux box